Neutralize mirror_structured_content wording: no model-vs-app routing claims

The opt-out is a wire choice (emit the serialized content copy or not); which
field a client forwards to a model is the client's decision, not the SDK's.
Reword the docstring, parameter help, and docs to describe duplication rather
than asserting content is "for the model."

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: olaservo

docs/servers/structured-output.md

@@ -236,17 +236,17 @@ There is one way to end up unstructured without asking for it: return a class th
 
 ## Skip the text copy
 
-By default a structured tool fills **both** channels: `structured_content` for the application, and a serialized copy in `content` for the model. That copy is the spec's recommendation (a SHOULD, not a MUST) so an old client that only reads `content` still sees the value. When the payload is large, or the host routes `structured_content` to the model itself, the copy is wasted — the same data crosses the wire twice.
+By default a structured tool sends the value twice: as `structured_content`, and as a serialized copy of it in a `content` text block. That copy is the spec's recommendation (a SHOULD, not a MUST): it keeps the value available to a client that consumes `content`. When the payload is large it is pure duplication — the same data crosses the wire twice. Which field a client actually forwards to a model is the client's choice; this flag only controls whether the serialized copy is emitted.
 
-Pass `mirror_structured_content=False` to return `structured_content` only, with empty `content`:
+Pass `mirror_structured_content=False` to send `structured_content` only, with empty `content`:
 
 ```python
 @mcp.tool(mirror_structured_content=False)
 def list_accounts(segment: str) -> list[Account]:
     return query_accounts(segment)  # only structured_content is sent
 ```
 
-The default is `True`, so nothing changes unless you opt out. A tool with no `output_schema` is unaffected — its `content` is the only representation and is always sent. If you want a *smaller* model-facing rendering rather than none, build the `CallToolResult` yourself and set `content` to a summary.
+The default is `True`, so nothing changes unless you opt out. A tool with no `output_schema` is unaffected — its `content` is the only representation and is always sent. If you want a *smaller* `content` rendering rather than none, build the `CallToolResult` yourself and set `content` to a summary.
 
 ## Recap
 

src/mcp/server/mcpserver/server.py

@@ -594,9 +594,9 @@ def add_tool(
                 - If True, creates a structured tool (return type annotation permitting)
                 - If False, unconditionally creates an unstructured tool
             mirror_structured_content: Whether structured output is also serialised into a
-                `content` text block (the spec's SHOULD, default True). Set False to return
-                `structuredContent` only -- useful when the host routes structured content to
-                the model itself and the serialised copy would double the payload.
+                `content` text block alongside `structuredContent` (the spec's SHOULD,
+                default True). Set False to send `structuredContent` only, without the
+                duplicate serialised copy on the wire.
         """
         self._tool_manager.add_tool(
             fn,
@@ -650,9 +650,9 @@ def tool(
                 - If True, creates a structured tool (return type annotation permitting)
                 - If False, unconditionally creates an unstructured tool
             mirror_structured_content: Whether structured output is also serialised into a
-                `content` text block (the spec's SHOULD, default True). Set False to return
-                `structuredContent` only -- useful when the host routes structured content to
-                the model itself and the serialised copy would double the payload.
+                `content` text block alongside `structuredContent` (the spec's SHOULD,
+                default True). Set False to send `structuredContent` only, without the
+                duplicate serialised copy on the wire.
 
         Example:
             ```python

src/mcp/server/mcpserver/utilities/func_metadata.py

@@ -116,14 +116,15 @@ def convert_result(
         flow surfaces on the wire as `resultType: "input_required"` rather than
         being JSON-dumped into a text block.
 
-        For a tool with an output schema, structured output is returned as
+        For a tool with an output schema, the structured output is returned as
         `structuredContent`. By default it is *also* serialised into a `content`
-        text block, so a client that reads only `content` still sees the value --
-        the spec's SHOULD. When ``mirror_structured_content`` is False the
-        serialised copy is omitted and `structuredContent` becomes the sole
-        representation; a host that routes `structuredContent` to the model itself
-        then no longer receives the payload twice. Because the serialised-text
-        guidance is a SHOULD (not a MUST), opting out stays conformant.
+        text block -- the spec's SHOULD, which keeps the value available to a
+        client that consumes `content`. When ``mirror_structured_content`` is
+        False that serialised copy is omitted and `structuredContent` is the only
+        representation on the wire, so the same payload is not sent twice. Because
+        the serialised-text guidance is a SHOULD (not a MUST), opting out stays
+        conformant. The SDK makes no assumption about which field a client routes
+        to a model versus elsewhere; that is the client's choice.
         """
         if isinstance(result, InputRequiredResult):
             return result